INFORMS 2021 Program Book

INFORMS Anaheim 2021

MB20

4 - Online Segmentation of Dynamic Data Stream Using Dynamic Classification Unit Model and Expandrogram Visualization Anna Khalemsky, Lecturer, Hadassah Academic College, Jerusalem, Israel, Roy Gelbard The incremental dynamic classifier DCU supports real-time segmentation processes in big and dynamic data environments. The model suggests using small data buffers, as an alternative to the reexamination of all past data for the updating of existing segments. To support the calibration of diverse domains, the model accommodates different forms of processing by using a wide range of parameters. The decision-making process strictly depends on the user’s preferences or implementation requirements. Comprehensive visualization, named ExpanDrogram, can potentially improve the interpretation of sophisticated segmentation processes and allow the user to participate in decision-making. MB20 CC Room 203B In Person: Healthcare Delivery General Session Chair: Sandeep Rath, University of North Carolina at Chapel Hill, Kenan Flagler, Chapel Hill, NC, 27599, United States 1 - Equity and Efficiency in Hospital Physicians’ Work Structure Masoud Kamalahmadi, University of Miami, Coral Gables, FL, United States, Kurt M. Bretthauer, Jonathan Eugene Helm, Alex Mills We study the fair and efficient assignment of patients to hospitalists (inpatient physicians) when hospitalists are partially localized to specific hospital units: each unit has a local hospitalist; however, hospitalists may be assigned to attend patients in the other units. We formulate a stochastic model of hospitalist-patient assignments and characterize the structure of the optimal policy that balances equity and efficiency in hospitalists’ work structure. We discuss how some of the policies that are commonly used in practice can be adjusted to achieve a better balance between equity and efficiency. 2 - Scheduling in Primary Care to Balance Patient Appointments and EHR Work Sandeep Rath, UNC Kenan Flagler Business School, Chapel Hill, NC, United States, Saravanan Kesavan, Bradley R. Staats Primary Care Physicians (PCPs) spend several hours a day working on Electronic Health Record Systems. EHR workload has been identified as a major source of PCP burnout. Broadly the PCPs have discretion on how to divide the daily EHR work. Currently, the appointment scheduling practices do not incorporate EHR workload when determining the daily appointment schedule for a PCP. Thus, PCPs have to manage the EHR work around the appointment schedule. We develop an optimization model which creates appointment schedules which explicitly incorporates EHR workload, as well as determines the best way to allocate EHR workload during the day. MB21 CC Room 204A In Person: Optimization Models in Healthcare General Session Chair: Oguzhan Alagoz, University of Wisconsin-Madison, Madison, WI, 53706-1539, United States Co-Chair: Elizabeth Scaria, University of Wisconsin-Madison, Madison, WI, United States 1 - Optimal Infection Control Interventions for Hospital Associated Clostridioides Difficile Infection: An Optimal Control Approach Elizabeth Scaria, University of Wisconsin-Madison, University of Wisconsin, Madison, WI, United States, Oguzhan Alagoz, Achal Bassamboo, Nasia Safdar Clostridioides difficile infection (CDI) is a common healthcare-associated infection. Hospitals use several controls simultaneously to mitigate CDI spread, including environmental cleaning and enhanced hand hygiene practices, but it is unclear how to select the best controls for a unique hospital. We formulate and solve an optimal control model using data from our agent-based model of CDI to characterize optimal infection control bundles in hospitals. We examine the optimal policy under varying budgets, disease distributions, and cost estimation methods. 2 - Novel Pooling Strategies for Genetic Testing, with Application to Newborn Screening Hussein El Hajj, Virginia Tech, Blacksburg, VA, United States, Ebru Korular Bish, Douglas R. Bish Cystic fibrosis is among the most prevalent life-threatening genetic disorders. For

cost-effectiveness, most cystic fibrosis screening processes start with a biomarker test, followed by a more expensive and accurate genetic test (DNA) for those newborns with elevated biomarker levels. To overcome the cost barriers of expanding genetic testing, we explore a pooled approach for DNA testing. This leads to a novel pooling problem that involves selection of variants for screening, potential partition of the selected variants into multiple panels, and selection of pool size for each panel. We establish key structural properties of optimal pooled DNA designs; develop an exact algorithm that generates a family of optimal pooled DNA designs, along with their corresponding budgets; and characterize the conditions under which a one-panel versus a multi-panel design is optimal. MB22 CC Room 204B In Person: Analytics for COVID and Pandemic Response General Session Chair: Dan Andrei Iancu, Stanford University, Stanford, CA, 94305- 7216, United States Co-Chair: Dragos Florin Ciocan, INSEAD, Fontainebleau, France 1 - Forecasting Covid-19 with Application to Vaccine Trial Design Michael Lingzhi Li, Massachusetts Institute of Technology, Cambridge, MA, 02111, United States, Hamza Tazi Bouardi, Omar Skali Lami, Thomas Trikalinos, Nikolaos Trichakis, Dimitris Bertsimas To help combat the COVID-19 pandemic and understand the impact of government interventions, we develop DELPHI, a novel epidemiological model. We applied DELPHI across >200 regions since early April 2020 with consistent high predictive power. DELPHI compares favorably with other models and predicted large-scale epidemics in areas such as South Africa and Russia weeks before realization. Furthermore, using DELPHI, we can quantify the impact of interventions and provide insights on future virus incidence under different policies. We illustrate how Janssen effectively accelerated the Phase III trial of the first single-dose vaccine Ad26.Cov2.S by selecting optimal locations using such analysis. 2 - Deploying a Data-driven Covid-19 Screening System at the Greek Border Vishal Gupta, University of Southern California, Marshall School of Business, Los Angeles, CA, 90026, United States, Hamsa Sridhar Bastani, Kimon Drakopoulos In the summers of 2020/2021 in the wake of the COVID-19 pandemic, many European countries sought to ease restrictions on non-essential travel to bolster its tourist economy, while still safeguarding public health. In collaboration with Greece, we deployed a data-driven COVID-19 testing system, which used real- time, bandit feedback to allocate scarce testing resources to I) identify asymptomatic, infected travelers ii) monitor different populations for potential spikes that merited adjusting border policies. We describe the system and document its effectiveness over summer of 2020; our data-driven system is able to prevent twice as many asymptomatic infections from entering the border than random, surveillance testing. We discuss some implications for the use of AI in managing the pandemic. 3 - COVID-19 Vaccine Allocation Optimization by Age and Risk Groups Vaccines are the primary means for mitigating a pandemic, but mass vaccination does not typically begin until a pandemic is well underway. As various types of COVID-19 vaccines become available in the US, it is crucial to decide on a vaccine prioritization strategy. We present an age and risk structured epidemiological model that incorporates vaccine allocation. We apply a derivative-free optimization algorithm as well as a greedy heuristic into our SEIR-type simulation model to determine an optimal vaccine rollout to minimize an objective, which can incorporate expected mortality, infections, and hospitalizations, accounting for both general ward and ICU beds. 4 - Quantifying the Benefits of Targeting for Pandemic Response Dan Andrei Iancu, Stanford University, Stanford, CA, 94305-7216, United States, Sergio Camelo-Gomez, Florin Ciocan, Xavier Warnes, Spyros Zoumpoulis The social-distancing measures implemented in response to COVID-19 have involved targeting specific groups or activities for confinement. Such targeting can be contentious, so rigorously quantifying its health and economic benefits is critical for designing effective and equitable policies. We propose a framework for computing interventions targeted by population group characteristics as well as the activities that individuals engage in, and showcase a full implementation using publicly available data. We find that optimized dual-targeted policies have a simple and explainable structure, and lead to substantial complementarities and Pareto improvements, reducing the overall number of deaths and the economic losses, and also reducing the time in confinement for each population group. Nazlican Arslan, Northwestern University, Evanston, IL, United States, Ozge Surer, David Morton, Lauren Meyers

50